Forecasting EEG time series with WaveNet
ObjectiveForecasting electroencephalography (EEG) signals, i.e., estimating future values of the time series based on the past ones, is essential in many real-time EEG-based applications, such as brain- computer interfaces and closed-loop brain stimulation. As these applications are becoming more and more common, the importance of a good prediction model has increased. Previously, mainly the autoregressive model (AR) has been employed for this task -- however, its prediction accuracy tends to fade quickly as multiple steps are predicted. We aim to improve on this by applying deep learning to make robust long-range forecasts.

MethodsWe applied the deep neural network model WaveNet to forecast resting-state EEG in theta- (4-7.5 Hz) and alpha-frequency (8-13 Hz) bands. We also compared WaveNet to the AR model, which has previously been widely used in real-time EEG applications.

ResultsWaveNet reliably forecasted EEG signals in both theta and alpha frequencies. It outperformed the AR model in estimating the signal amplitude and phase.

ConclusionWe demonstrate for the first time that deep learning can be utilised to forecast resting-state EEG time series over 100 ms ahead.

SignificanceIn the future, the developed model can enhance the real-time estimation of brain states in brain-computer interfaces and brain stimulation protocols. It may also be useful for answering neuroscientific questions and for diagnostic purposes.